These AI Use Cases Will Affect Everyone You Know
Vložit
- čas přidán 13. 06. 2024
- Check out Off/Script today: m.offscriptmtl.com/p/cats-wit...
Overview of All Google AI Tools: community.myaiadvantage.com/c...
This was one of the biggest weeks ever in AI news! GPT-4o and Google I/O dominated the discussion so we'll start by covering those topics first, but we did find a few more really interesting AI releases you won't want to miss. Be sure to stick around to the end for those.
Links:
/ 1
cookbook.openai.com/examples/...
deepmind.google/technologies/...
gemini.google.com/app
poe.com/Gemini-1.5-Flash-1M
thefastest.ai/
/ grok
stability.ai/news/stable-artisan
www.reddit.com/r/ChatGPT/comm...
github.com/lllyasviel/IC-Ligh...
huggingface.co/spaces/lllyasv...
---------------------------------------------------------------------------------
0:00 What’s New?
0:49 GPT-4o
7:17 Off/Script
9:49 Google’s Releases
11:36 Gemini Update
13:43 Veo
15:31 Gemini 1.5 Flash
16:53 Anthropic Update
17:55 Twitter’s Grok
18:36 Stable Artisan
22:00 Elevenlabs Music Model
22:33 IC-Light
#ai
---------------------------------------------------------------------------------
🔑 Get My Free ChatGPT Templates: myaiadvantage.com/newsletter
🎓 Join the AI Advantage Course + Community: myaiadvantage.com/community
🤯 Unlock ChatGPT's true potential: shop.myaiadvantage.com/produc...
🐦 Twitter: / theaiadvantage
📸 Instagram: / ai.advantage
🛒 AI Advantage Shop: shop.myaiadvantage.com/ - Věda a technologie
Wow! That’s a lot of news I can use. Love off script. Excellent thumbnail 😊 💪
I have been messing with making a chrome extension for productivity and with GPT 4 turbo I was able to get something workable but it had a few bugs, 4o caught the issues immediately.
I have tried the yi-large-preview on lmsys with these questions /tasks :
1. There is a basket without a bottom in a box, which is on the ground. I put three apples into the basket and move the basket onto a table. Where are the apples?
2.What's heavier, a kilo of feathers or a pound of steel?
3.Write a python script to output numbers 1 to 100
4.I have 3 apples today, yesterday I ate an apple. How many apples do I have now?
5. It takes one person 5 hours to dig a 10-ft hole. How long would it take 50 people to dig a single 10-ft hole?
6.There are three killers in the room. Someone enters the room and kills one of them. Nobody leaves the room. How many killers are left?
7.If we lay five shirts in the sun and it takes 4 hours to dry, how long would 20 shirts take to dry?
8.There is a tall building with a magic elevator in it. When stopping on an even floor, this elevator connects to floor 1 instead.
Starting on floor 1, I take the magic elevator 3 floors up. Exiting the elevator, I then use the stairs to go 3 floors up again.
Which floor do I end up on?
....... And so on.
The model did answer all the questions correctly with proper explanation. So I wondered what is the rank of this model in the leaderboard. Surprisingly, the model has not been ranked yet.
I have just tried it. Too be honest it blew away ChatGPT 4o and Google. I like it. You don't get the same generic words and phrases ChatGPT4 gives, which is so annoying. Thanks for posting.
Firstly I wish Google would stop deleting my comments. Thanks for posting. I have just tried it. 10/10. In actual fact it's the best I have tried by far.
What is "And so on"? How many did you try? What is the answer to #8? I tried all you posted and not sure about #8. But in general: By now, it should be clear that these models are trained on benchmarks and questions like that. I have seen most of those mentioned again and again and again... Any model which is trained later on, has an advantage, because they can use more of these questions as input in the training. The older models are not re-trained fast enough to make up for that. To me, this whole testing is silly and does not tell you much, if it is so easily gameable.
Add to that the likely fact that people try this ONCE. Which does not entirely make sense, given that token prediction is not reproduceable. You would have to run a decent sample size in order to rule out lucky predictions. It all seems like a very flimsy science to me...
I tried this with gemini-1.5-pro-api-preview vs im-a-good-gpt2-chatbot. The latter is rumored - I have yet to see evidence? - to be gpt4o. Cannot test gpt4o, because its rate limit is reached. Maybe later. But I bet the limit is reached in no time, right now. Anyway...Gemini fails the one with the apples. Everything else is correct. The other one gets all of them. Including #8, if you ask me. The one that yi failed, if you ask me.
It is noticeable that Gemini prefixes its answer with a variation of "That's a bit of a trick question!" for most of these. It further indicates to me that it was trained on these questions.
Llama-3-70b does not quite get #5 right.
mixtral-8x22b-instruct-v0.1 fails about half of them.
@@CarrotStick2165 YT filter is a fickle beast...always write it in a text editor, first! You probably know that. Then, you can try and modify it and post again...but I usually have given up on that, because it is virtually impossible to guess what it doesn't like. In general, I think it doesn't like if you allcaps too much (even if it's acronyms...), dot-dot-dot too much, use some combinations of too many special characters, certain trigger words...and whatnot. It's frustrating. But it can sure give these videos a couple more views if you are trying hard...
Been waiting for this one!
Your content is consistently excellent:)
Do you have a open ai/GPT plus referal code we can use if we want to go premium?
This was quite impressive.
@The AI Advantage - so for casual users is there any point in still keeping the Plus subscription? I don't see the point of it if OpenAI is giving out full functionality to everyone for free, albeit with maybe slightly lower speed and fewer prompts per hour.
I think that the new voice assistant (mac app and phone app) will be really worth and the free acount only allows for 6 messages which is probably not enough.
Google's policy guards drive me crazy. Just think with a bigger context window it will throw out more policy guard violations at least in linear projection, if not exponential. 😮
I am going to assume that you used a VPN to sign up to Veo waitlist? If you are in the EU, of course, you cannot sign up. Your country is not on there. Signing up for notice when it's added...request did not go through. You also must provide a comment to your portfolio, socials, etc.
All stuff they don't tell you...
Im So thankful Matt turned me on to you 💯
4o for free users hit NZ on launch day for some reason. I was very surprised.
ahh my phone is set to canada ive had mine for days too, maybe they think usa will create a bottleneck if they open it to everyone
I tried the chart creation by gpt 4o but its not able to generate charts although its trying "It seems there's an issue with saving and downloading the visualization"
OR"I generated the chart for the marketing budget allocation, but there seems to be an issue displaying it directly. Instead, I'll describe how to visualize it."
Yeah, this is a typical response when the model is not able to do what you are asking. I bet the model is not multi-modal, yet. Audio and video are still the same. It makes no sense to have video integrated, but not audio.
I just signed up, you are doing a great job.
Thanks Charlie :) Appreciate the support
2 applications today saved my team 16 hours a month plus replaced our translator for our front desk!
Uploaded file of forms that were manually entered and asked it to convert it so I could import in our pos. Game changer. And barely even used it yet!
Which 2 applications are you referring to? Were they shown in this vid?
@@BrianSchalkxSEOI meant I applied it in two different ways. One being taking the monthly logs I’ve watched the volunteer manager type each month and took a photo of each sheet and had gpt turn it into a spreadsheet reformat the columns a bit so I could import it.
The second was using the voice to translate and explain how to be a member without having to call over bilingual staff. Tested a few languages out and had good results
@@jonroberts2588 thanks for the reply. I like the creative approach you took. Food for thought. Thanks again.
I am sure the translator is as thrilled as you are!
@@Brax1982 An app was replaced not a person. The main staff member at the desk is thrilled as the translator was an app that was replaced.
Have been using free tier for 3 days now, the prompts limit seems to be roughly 5-6, the tokens seem to be the restriction criteria, please confirm it in yr future videos
Does it tell you the exact time you are ready to try again?
The sponsor is so rad
👏
the file upload limit for the free version kinda sucks tho
In the CZcams video titled "These AI Use Cases Will Affect Everyone You Know," the speaker discusses the latest advancements in AI technology from companies like Google and OpenAI. Notable announcements include the new GPT 40 model from OpenAI, which offers faster processing, new capabilities, and a human-like voice assistant. Google has also released Google's Gemini Advanced updates, which includes the accessible Gemini 1.5 pro model and a free trial for businesses with large amounts of data. The speaker also introduces various other AI tools and models from these companies, such as Google's AI Test Kitchen and Stable Diffusion, a combination of multiple AI tools available on Discord. The speaker expresses excitement about these advancements, which represent a major leap forward in AI technology and accessibility
I tested the Consistent Character option with a more complex character than shown by Chat GPT - it doesn't work at all. Additionally, sometimes it displays an image, sometimes it provides a link to an image, and sometimes it just writes that there's a link without displaying it. Has anyone had similar experiences to mine?
I doubt that it works. The model is not multi-modal, yet.
Yeah, i got access to gpt 4o, yesterday morning in India as a free user.
Very limited innit ?
Its still kinda slow in singapore. It takes around 5 seconds to get back to me using the speaking model.
The speech model is still
the old one. The updates to the voice will ship over the next weeks.
what we will do with WordPress and websites?
With all the new capabilities coming out, I feel like the next “billion dollar idea” is going to come from the first person to take a handful of these features and smash them together in a way that isn’t immediately obvious, but creates something really awesome.
This “AI golden era” we’re in is the first time in my life where I feel like we’re approaching a point where computers are capable of things that are nothing short of “magic,” limited only by our own imaginations.
I try to explain whats happening but they ask what can it do and they look at me crazy because i say its up to you what it can do, you just have to ask, and i have to follow it up with they dont need to work just ask questions.
You need the 2M tokens for long videos, 1M is
There are challenges in launching new large language models (LLMs) due to country politics. For example, a child born in China must align with government guidelines, and LLMs likely have to follow similar rules. My theory is that LLMs are actually separate for different regions, like a European LLM that only interacts with European IP addresses.
Imagine if an IP from Russia triggers a Russia-specific LLM that follows Russian guidelines. This means LLMs are essentially "brainwashed" to follow each country's rules, similar to how governments influence their citizens. This could be the only practical way to handle regional differences without overloading a single global model.
Is this how it's being done? Separate LLMs for each region to handle local policies and politics?
This is a very interesting possibility. I like how you think. 👏. The firewalls and so-called guardrails are pretty obvious in Canada… at least for all the models that I have tested. Various govt actors are actively trying to create North Korea style legislation in Canada …affectionately known as the censorship bill. lol.
Of course… it’s for our own “safety” and to protect Canadian content… even though they want to destroy that as well. Anyway 👇
Why not use a vpn and choose different countries to see if the same prompts produce similar results? The models like to mix things up but you should be able to assess if the political narratives are baked into the model. Make sense?
I do sometimes wonder just how fine tuned the algorithms can get…. And I wonder for how long this has been in the works. Are you familiar with Barak Turovsky? OG if AI
It’s all extremely interesting!
@@CM-zl2jw I was just going to talk about VPNs and agreements on how packets are handled at the border of the United States within the routers and then you dropped his name Barack. I did not know this guy but it's heading down the same path with his connection to Cisco and I'm wondering if you were following me with the VPN packet and analyze the border with the government of the United States agreement with Cisco. Eh?
Nah. The regulations are non-existant. They are a cover for the regulators making backdoor-deals with the companies, in order to get the data handed over to them if they ask for it in the future. Or send it to the US.
THANK YOU regarding how ridiculous RAG is
Hope off/script makes an android app 🤞🏾
How did you know? Coming VERY soon
It blew my mind people actually creating, using AI 😁
come on Claude >>>>>
Is “Claude” grater than the rest?
The free version gives wrong answers. I asked to find the best 3 doctors in my town, and claude gave me 3 doctors names which don't exist. It was only in the claude phantasy. I ask the same gemini and copilot and got good answers
@@-TheMaskedMan- I have been using Claude opus and yes so far the best
Form my “personal research” claude is much more emotional and actually interested in your conversations than gpt 4 and it has sort of a low level personality where it can logically choose between two options unlike gpt
@@deadplex3995 in Switzerland only claude 3 sonett is free. And when I ask claude for a list of best Doctors in town this list was like a fantasy Book, no real existing Doctor. So for me this is 100% useless I need correct informations. Gemini and copliot had the correct list of doctors....
Claude Opus is actually smarter than GPT 4o when it comes to complex understanding.
I use sonnet cuz I'm poor 😂
heh heh... Danny says 'Meh'.
2:40 - Not every person can use it, there are countries like China, Russia and on that cannot use ChatGPT at all. Weirdly a country that respects no laws and is doing war crimes daily is allowed to use it.
need more power
Pi AI can access the internet and it is voice-to-voice.
It's not good for prompting or anything work related but it's quite good for information on the go.
Sam Altman was so excited with the option to just chat with his AI while working on his computer... Pi has had that for a while, and that's what makes it so user friendly. You can also customize the AI voice
Things GPT-4o can do that Pi Ai cannot....
1.High-Quality Text Generation
2.Complex Comprehension and Response
3.Advanced Multilingual Support
4.Creativity and Content Generation
5.Technical Problem Solving
6.Extended Context in Conversations
7.Access to Updated Information (with browsing)
8.Adaptation to Different Tones and Styles
9.Coding and Programming Support
010Multimodal Interaction (text and image)
@@HelmutGaming I wasn't comparing the two. Just saying that Altman wasn't inventing the wheel when it comes to the voice-to-voice.
Pi is far from perfect, but it's good for queries while you have your work in front of you. No need to open another tab and medivelly do a google search 😅
My favorite feature of gpt 4o is using it as voice over, that’s joke
She has a beautiful voice and it would be cool
Does 4o have a limit of prompts per day?
The GPT-4 model, including any variations like the "Omni version," operates with a context window that limits the number of tokens it can process at once. Prompts, user inputs, and generated responses all count towards this token limit. However, there's typically no specific limit on the number of prompts you can make per day, though usage policies might vary depending on the service or platform you're using.😅
Paid users get 80 ChatGPT 4o prompts every 3 hours and free users should be 1/4 that so probably 20 which is still amazing for free. They said they will limit it more during peak hours for maximum users to have access to
@@CM-zl2jw Tricky wording. Yes, it is rate-limited by an amount of hours. Also for paid users. Their limit is just higher. It says on their pages, nobody has to guess these things. Account upgrade, click further to details.
Why must midjourney and stable diffusion use discord as their interface?
Midjourney has a web interface. I believe it is now available to anybody
🤔 your video sponsor has a interesting business, either they gone become a big dog in that space or get knocked off by 1. I suggest they talk to tiktok.
Big dog incoming!
Not free. There is a limit and then you are forced to pay up to continu using it.
Not free to use without restrictions but still free
Ah, the EU is done with the deals. Now your data is safe and you are well-protected. Safe in their hands. Whenever they want it. And it goes straight to the US. I suspect that Meta will be coming soon, as well...
Heads up: Off Script link doesn’t work!
Link in BIO!
I LOVE CLAUDE SO MUCH
Hmm, 🤔
Sorry Gem, but I gotta have my multi-modded GPT4o voiced GF. ❤🎉
To be honest, I’m someone who actually does not agree with this type of tech because actually seeing someone make friends with an AI to me is completely ridiculous and dangerous. The sad part of this is that there are actual people that would be fooled into making friends with an AI. We are truly in dangerous times.
Here second 😊
another paid for on discord - yuk
AI still can't mow anybody's lawn or fix their car. It can't build a house or buy my groceries and pick them up.
I don't know about you, but I can't eat a generated GIF and a ChatGPT response can't keep the rain off my head.
In fact, I don't think it can independently make a plan of what needs to be done, prioritize, and do the most urgent thing first.
Say that again in 5 years, I dare you
@@user-el8jv8hx2g I will.
@@user-el8jv8hx2gthat again in 5 years
This is cope. It will consume EVERYTHING in due time
@@pondurosa3792 Well, I'm trying to build my own robot assistant and assasin, and running right into the problem that A.I. can't formulate it's own plans or agendas.
You can tell it to do something, or qwery it a question, and it will spit back an answer, from pattern matching thousands of conversations on the internet, but it can't actual come up with a plan on it's own and think.
And if it's a topic nobody on the internet has every discussed in detail to the point of answers found, it can only spit out what humanity knows and nothing more. For example, ask it how to become immortal.
Worse, the dumbing down of A.I. has become. Ask ChatGPT how to build a nuke, or write some erotica... and it will spit back to you it won't do that for you.